KR20190109599A - Generating output for presentation in response to user interface input, where the input and/or the output include chatspeak - Google Patents

Abstract

Some embodiments herein relate to converting an Internet language into a canonicalized (standardized) format, where it is provided to a user through a user interface input device of the computing device, such as input provided by the user to an automated assistant. It is included in the natural language input represented by The canonicalized form of the Internet language may be used by an automated assistant to determine response content responsive to natural language input, and the response content may be provided to the user via one or more user interface output devices of the user's computing device. . Some implementations, instead of the normalized form of the Internet trendy language, are based, at least in part, on “Internet Vocabulary Measurements” determined based on past usage of the Internet language by the user and / or additional users for presentation to the user. It is further and / or alternatively to provide natural language output comprising an Internet idiom.

Description

Generate output to be presented in response to a user interface input where the input and / or output includes an Internet grammar {GENERATING OUTPUT FOR PRESENTATION IN RESPONSE TO USER INTERFACE INPUT

This disclosure relates to generating output for presentation in response to user interface input.

Many users can communicate with each other via text message exchange threads, which can be implemented using various technologies such as text message exchange, online chat (e.g. instant messenger, Internet relay chat, etc.), social network threads, and so on. have. Many users utilize so-called "chatspeaks" in various communications with other users, particularly through message exchange threads, when using mobile computing devices. The Internet word may include initialization (e.g., an abbreviation and / or abbreviation consisting of multiple words of initials (e.g., 'laughing loud'), and removing characters from the words (e.g. "review" as "rvw". "Really" as "rlly"), adding letters to words (e.g. "hello" as "hellooooooooo", "yes" as "yeeesssss"), may not appear in words It can take various forms, such as a word representation using one or more letters (eg "see" as "c" and "thanks" as "thx").

Some embodiments of the present disclosure relate to converting an Internet language into a normalized (standardized) format, where the Internet language is used, for example, of a computing device, such as a natural language input provided by an user to an automated assistant. Included in natural language input formulated by the user through the user interface input device. The canonicalized form of the Internet language can be used by an automated assistant in determining response content responsive to natural language input, and the response content can be provided to the user through one or more user interface output devices of the user's computing device. In many implementations, natural language input, including Internet language, can be formulated by the user as part of a message exchange thread between the user, automated assistant, and optionally one or more additional users. In some of these implementations, the response generated by the automated system based on the canonicalized form of the Internet language may be integrated into the message exchange thread. For example, the response may be inserted in the transcript of the message exchange thread displayed in the graphical user interface of the message exchange client operating on the user's computing device.

Translation of the Internet-facing language in user-provided inputs, for example, allows the user to interact with computing devices such as automated assistants, even with limited written language skills, because they communicate in languages that are dyslexia or not fluent. In other words, the computing device can determine the input even if the user cannot write / enter the query using the correct spelling and grammar. This may be particularly useful for computer devices that receive input in languages other than pronunciation, with English and French being examples of such languages.

As one example, some implementations of this specification can be utilized to convert natural language input of "cu I8er" to a normalized form of "see you later", where the normalized form is "see you then", "OK. It may be used by the automated system to determine the response content that is an appropriate "chat response" such as ". As another example, some implementations may be utilized to determine whether natural language input consisting solely of "thx" should be converted to the normalized form "thanks" used by the automated assistant, wherein the automated assistant is one or more specific terms. It may not be appropriate to use a normalized form in other natural language inputs containing "thx" with. For example, in natural language input including "thx company", "thx", "thx" cannot be translated as "thanks". Instead, the automated system can utilize "thx" to determine an appropriate response to "thx company", such as "search response" that includes content related to audio / video company THX.

Some embodiments herein may additionally and / or alternatively include a "chatspeak measure" determined based on past usage of the Internet word by additional users and / or users for presentation to the user. Based, at least in part, on providing a natural language output that includes the Internet language instead of the normalized form of the Internet language. For example, past instances of a pandemic translation for a user may be stored, and a pandemic measure may be determined based at least in part on the amount of the instance (eg, raw count, frequency and And / or other functions of amount). For example, if a user continues to use an internet language in a message provided to an automated assistant, that user's internet measurement may indicate that the user wants to receive a response from an automated assistant that includes the internet. The automated assistant may provide at least responses that include Internet language instead of the normalized version of the terms. Internet language measurement may be a "generic" Internet language measurement or may be applied to one or more specific groups of Internet language. For example, a user's Internet measure for "thx" / "t hanks" might be measured by an automated assistant in response to that user, "thanks" to "thx", or "gr8" / "great" A user's separate Internet Vocabulary measure for (or lack of the Internet Vocabulary measure for "gr8" / "great") may not have the automated assistant replace "great" with "gr8" in response to that user.

In some implementations, at least one of the one or more Internet language measurements used to determine whether to include the Internet language in a given response to a given user can be a particular measurement for a given user. For example, it may be determined based at least in part on instances of the Internet language of the message formulated by the user, such as instances of the Internet language recognized in accordance with the techniques described herein. In some of these implementations, Internet language measurements specific to a user may be specified for one or more devices of the user. For example, the first Internet measurement may be assigned to the user for the user's mobile computing device and the second internet measurement may be assigned to the user's "desktop" computing device. In some implementations, at least one of the one or more internet language measurements used to determine whether to include the internet language in a given response to a given user may be based on the plurality of internet language instances of the user. For example, a user having one or more attributes in common (eg, geographic area, age range) and / or a user that can be specific to the type of computing device used and / or the type of message exchange client used. May be specific to a group of users (eg, a message exchange client A of a plurality of users may have a greater frequency of Internet traffic than client B).

The natural language input of the user may be received and processed in various scenarios and / or the natural language output generated by the automated system may be generated in various scenarios. For example, natural language input may be input provided by a user during communication with one or more other users, such as communication via chat, SMS, and / or other message exchanges. As another example, the natural language input may be provided to an automated assistant participating in a conversation with a user via one or more user interface input and output devices. For example, the automated assistant may be integrated in whole or in part on a user's computing device (e.g., mobile phone, tablet, device dedicated to automated assistant functions), and one or more user interface input devices (e.g. microphone) , A touch screen) and one or more user interface output devices (eg, a speaker and a display screen). Also, for example, an automated assistant may be implemented in one or more computing devices that are in communication with, but in whole or in part with, a user's client computing device.

In some implementations, a method is provided that is performed by one or more processors, the method comprising determining an Internet language measurement for one or more users based on a plurality of messages that contribute to a message exchange thread by one or more users. Include. The method further includes receiving natural language input submitted by one of the users as part of an ongoing message exchange thread between the user and the automated assistant. The natural language input is formulated by the user through a user interface input device of the user's computing device. The method further includes generating a natural language output for inclusion in a response of the automated assistant to the natural language input. Generating the natural language output includes incorporating one or more internet words into natural language output based on the internet language measurement. The method further includes including a response of the automated assistant in the message exchange thread, the response comprising a natural language output having one or more Internet languages.

Each of the present methods and other implementations of the technology disclosed herein can optionally include one or more of the following features.

In some implementations, including the response in a message exchange thread includes inserting the response in a transcript of a message exchange thread displayed in a graphical user interface of a message exchange client operating on a user's computing device. .

In some implementations, the Internet Vocabulary measure is specific to one or more Internet Vocabulary term groups that include at least one or more Internet Vocabulary words included in the natural language output, and incorporating the one or more Internet Vocabulary terms into the natural language output includes: It is based on internet language measurement specific to the group containing the buzzword.

In some implementations, incorporating the Internet language into a natural language output includes identifying an initial natural language output responsive to the natural language input; Determining whether one or more predetermined terms of the initial natural language output map to the Internet trendy language; And substituting certain terms of the initial natural language output with an Internet fashion.

In some implementations, the method further includes determining additional internet language measurement for one or more of the users, wherein generating the natural language output is based on the internet language measurement and based on the additional internet language measurement. Includes steps to include one or more Internet words in the natural language output. In some of these embodiments, the internet measure is applicable to a plurality of internet words that are user specific and include one or more additional internet words that are not integrated into the natural language output. In some versions of these implementations, additional Internet language measurements are not user specific and are specific to at least one of the Internet languages integrated in the natural language output.

In some implementations, Internet language measurements are user specific. In some of these implementations, the message exchange thread includes at least an automated assistant message exchange thread between the user and the automated assistant. In some versions of these implementations, determining the internet language measurement is based on a count of instances in which the internet language is used in the messages, and the method further comprises determining one of the instances on which the count is based. Include. A method of determining one of the instances comprises receiving a predetermined one of the messages in one of past automated assistant message exchange threads, wherein the predetermined message is formulated by a user and at least the first term and the first. Contains 2 terms-; For at least a first term of the natural language input, identifying a first normalization of the first term; Generating a candidate alternate input for the natural language input, wherein the generating replaces the first term with the first normalization and a second normalization or a second normalization of the second term in the candidate substitute input. Including the term-; Determining a first score for the natural language input; Determining a second score for the candidate substitute input; Selecting the candidate substitute input based on the first score and the second score; Determining response content based on the candidate substitute input; Incorporating the response content into one of past automated assistant message exchange threads as a response to natural language input by the automated assistant, wherein incorporating the response content is based on selecting the candidate alternate input. -; And storing an indication of one of the instances based on selecting the candidate replacement input.

In some further versions of these implementations, determining a first score for natural language input is based on applying at least a first term and a second term to a language model and determining a second score for the candidate substitute input. The step of doing is based on applying at least one of the first normalization and the second term or the second normalization to the language model. In another version of these implementations, the internet measure is specific to the user, is specific to the internet, and storing an indication of one of the instances based on selecting the candidate alternative input Storing the association of the indication to at least the first term or the first normalization.

In some implementations, determining the internet language measurement includes: a count of instances in which the internet language is used in the messages, an amount of one or more internet language terms used in the messages, commonality of the internet language terms, and / or Or based on consistency in the use of the Internet idioms.

In some implementations, a method is provided that is performed by one or more processors, the method comprising receiving natural language input submitted by a user as part of a message exchange thread between at least the user and an automated assistant. The natural language input includes at least a first term and a second term and is formulated by the user through a user interface input device of the user's computing device. The method includes identifying, for at least a first term of the natural language input, a first normalization of a first term; And generating a candidate substitute input for the natural language input. Generating the candidate substitute input includes replacing a first term with a first normalization and including a second term or normalization of the second term in the candidate substitute input. The method includes determining a first score for the natural language input based on applying at least the first term and the second term to a language model; Determining a second score for the candidate substitute input based on applying either the first normalization and the second term or the second normalization to the language model; Selecting the candidate substitute input based on the first score and the second score; Determining response content based on the candidate substitute input; And incorporating the response content into the message exchange thread as a response to the natural language input by the automated assistant. Including response content in the message exchange thread is based on selecting the candidate replacement input.

Each of the present methods and other implementations of the technology disclosed herein can optionally include one or more of the following features.

In some implementations, including the response content in a message exchange thread includes inserting the response content into a transcription of a message exchange thread displayed in a graphical user interface of a message exchange client operating on the user's computing device. .

In some implementations, the method further includes determining a response content score for the response content and determining the first score further based on the response content score. In some of the embodiments where a response content score is determined, the response content is a scripted response mapped to one or more input parameters, and determining the response content score comprises at least a portion of the input parameters and the Based on the comparison of candidate replacement inputs. In some of the embodiments in which a response content score is determined, the response content is based on one or more search results responsive to a candidate replacement input, and wherein determining the response content score is characteristic of at least one of the search results. Or based on the score. In some versions of these implementations, determining the response content score is based on at least one characteristic of the search results indicating that the search result is a specific type of search result.

In some implementations, the language model is an n-gram model and uses the first score for the natural language input based on applying at least the first term and the second term to the language model. Determining includes applying both the first term and the second term to the language model.

In some implementations, the method further includes generating the language model based on normalization of a message from a message exchange thread between humans, wherein the normalization of the message is such that the chatspeak terms are Internet-interpreted. When used as a chatspeak, it is a qualified term that replaces the Internet language of these messages.

In some implementations, determining response content based on the candidate replacement input includes searching one or more response resource databases based on the candidate replacement input.

In some implementations, a method is provided that is performed by one or more processors, the method comprising receiving natural language input submitted by a user as part of a message exchange thread between at least the user and an automated assistant. The method includes, for at least a first term of natural language input, identifying a first normalization of the first term; And generating a candidate substitute input for the natural language input.

Generating the candidate substitute input includes including a first normalization in place of the first term in the candidate substitute input. The method includes determining a first score for the natural language input based at least on applying the first term to a language model; Determining a second score for the candidate substitute input based on applying at least the first normalization to the language model; Selecting the candidate substitute input based on the first score and the second score; Determining response content based on the candidate substitute input; And incorporating the response content into the message exchange thread as a response to the natural language input by the automated assistant. Incorporating response content into the message exchange thread is based on selecting the candidate replacement input.

Further, some implementations include one or more processors of one or more computing devices, wherein the one or more processors are operable to execute instructions stored in an associated memory, wherein the instructions are configured to perform any of the methods described above. Some implementations also include a non-transitory computer readable storage medium that stores computer instructions executable by one or more processors to perform any of the methods described above.

It should be understood that all combinations of the foregoing concepts and additional concepts described in more detail herein may be considered as part of the subject matter disclosed herein. For example, all combinations of claimed subject matter appearing at the end of this specification are considered to be part of the subject matter disclosed herein.

1 is a diagram of an example environment in which implementations disclosed herein may be implemented.
Figure 2 receives a natural language input including the Internet word, converts the Internet word into a normalized form, generates a response content in response to the natural language input using the normalized form, presented in response to the natural language input An example of providing response content to do is shown.
3 illustrates an example client computing device having a display screen displaying an example of a conversation that may occur between a user of the client computing device and an automated assistant in accordance with implementations described herein.
4 illustrates an example client computing device having a display screen displaying an example of a conversation that may occur between a user of the client computing device and an automated assistant in accordance with implementations described herein.
FIG. 5 illustrates the use of normalized forms to receive natural language input including the Internet language, convert the Internet language into a normalized form, and generate response content responsive to the natural language input, A flowchart illustrating an example method of providing response content to present in response.
FIG. 6 determines an Internet language measurement for at least one user, receives a natural language input from the user, determines a natural language output based on the internet language measurement and the natural language input, and presents to the user in response to the natural language input. A flow diagram illustrating an exemplary method of providing natural language output for display (shown) is shown.
7 illustrates an example architecture of a computing device.

Users may wish to utilize chatspeak when submitting requests to various non-human computer systems, such as message exchange threads with automated assistants. For example, users may be accustomed to using the Internet language language and / or the constraints of the user's computing device (eg, display constraints, user interface input device constraints) may be preferred to use the Internet language. However, many computer systems may not recognize the Internet language, may not properly interpret it in a normalized form, or may misinterpret the input in the Internet even if the input is not intended. In addition, many computer systems may not respond to an Internet fashion and / or may not respond to an internet fashion based on past usage of the internet in a message exchange thread by one or more users.

Embodiments of the present disclosure relate to converting an Internet language into a canonicalized form, where it is formulated by a user through a user interface input device of a computing device, such as natural language input provided to an automated system by a user. Are included in the natural language input. The canonicalized form of the Internet language may be used by an automated system to determine response content responsive to natural language input, which response content may be provided to the user through one or more user interface output devices of the user's computing device. .

As one example, some implementations of this specification can be utilized to convert natural language input of "cu I8er" to a normalized form of "see you later", where the normalized form is "see you then", "OK. It may be used by the automated system to determine the response content that is an appropriate "chat response" such as ".

Some embodiments herein may additionally and / or alternatively include a "chatspeak measure" determined based on past usage of the Internet word by additional users and / or users for presentation to the user. Based, at least in part, on providing a natural language output that includes the Internet language instead of the normalized form of the Internet language. For example, past instances of a pandemic translation for a user may be stored, and the pandemic measure may be determined based on the quantity of that instance (eg, raw count, frequency and And / or other functions of quantity). For example, if a user continues to use an Internet word in a message provided to an automated assistant, the user's measurement of that word's Internet word is measured by that user's Internet word (or some specific chatspeak terms). It may be indicated that it wants to receive a response from an automated assistant that includes it, and the automated assistant may provide at least responses that include chatspeak terms instead of the normalized version of the terms. Another basis for determining Internet language measurements is the quality of Internet terminology used (e.g., complexity), and the commonality of the Internet terminology used (e.g. "laugh out loud" as "lol"). It is common to substitute for a particular Internet term for a form), and the user's consistency of use of a particular Internet term (eg, a user uses a particular term for most of the time instead of the normalized form). Indeed, Internet language measurements may be determined from combinations of these attributes (weighted or otherwise) for the general response, including the use of the specific internet term in the Internet word and / or response.

Referring to FIG. 1, an example environment in which the techniques disclosed herein may be implemented is shown. An example environment includes a plurality of client computing devices 1061 -N and automated assistant 120. Although automated assistant 120 is shown in FIG. 1 separately from client computing device 1061-N, in some implementations all or aspects of automated assistant 120 are one of client computing devices 1061 -N. It can be implemented by the above. For example, client computing device 106i may implement one or more instances of automated assistant 120, and client computing device 106N may also implement separate instances of one or more aspects of automated assistant 120. Can be. In an implementation where one or more aspects of automated assistant 120 are implemented by one or more computing devices remote from client computing device 1061 -N, the client computing device 1061 -N and automated assistant 120 may be referred to. Aspects may communicate over one or more networks, such as a local area network (LAN) and / or a wide area network (WAN) (eg, the Internet).

The client computing devices 1061 -N may be, for example, desktop computing devices, laptop computing devices, tablet computing devices, mobile phone computing devices, computing devices of user vehicles (eg, in-vehicle communication systems, in-vehicle entertainment systems, vehicles One of the user's wearable devices (eg, a user's watch with a computing device, a user's glasses with a computing device, a virtual or augmented reality computing device), including a navigation system), a personal assistive computing device, or a computing device. It may contain the above. Additionally and / or alternatively, a client computing device may be provided. In some implementations, a given user can communicate with automated assistant 120 using a plurality of client computing devices of the user that collectively form a coordinated “ecosystem” of the computing devices. However, for brevity, many examples described in this disclosure will focus on a given user who manipulates one of the client computing devices 1061 -N.

Each client computing device 1061 -N may operate a variety of different applications, such as a corresponding one of the message exchange clients 1071 -N. The message exchange client 1071 -N may exist in various forms, the form may vary across the client computing device 1061 -N, and / or multiple forms operate on one of the client computing devices 1061 -N. Can be. In some implementations, the one or more message exchange clients 107I-N can be a short messaging service ("SMS") and / or multimedia messaging service ("MMS") client, an online chat client (eg, instant messenger, Internet relay). Chat, or " I RC ", etc.), messaging applications associated with the social network, personal assistant messaging services dedicated to conversations with automated assistant 120, and the like. In some implementations, one or more message exchange clients 1071 -N may be implemented via a web page or other resource rendered by a web browser (not shown) or other application of client computing device 106.

As described in more detail herein, automated assistant 120 receives input from one or more users and / or one or more users via user interface input and output devices of one or more client devices 1061 -N. Gives output. In some implementations, the user interface input described herein is explicitly passed to the automated assistant 120. For example, one of the message exchange clients 107I-N may be a personal assistant messaging service dedicated to a conversation with the automated assistant 120, and the user interface input provided through the personal assistant messaging service may be an automated assistant ( 120 may be automatically provided. Also, for example, the user interface input is explicit to the automated assistant 120 of one or more message exchange clients 1071 -N based on a particular user interface input indicating that the automated assistant 120 will be invoked. Can be delivered. For example, certain user interface inputs can be one or more typed characters (such as @PersonalAssistant), user interactions with virtual buttons (such as tabs, long tabs), verbal commands (such as "Hey Personal Assistant"), and so on. have. In some implementations, the automated assistant 120 can perform one or more actions in response to the user interface input even if the user interface input is not explicitly passed to the automated assistant 120. For example, automated assistant 120 may examine the content of the user interface input and perform one or more actions in response to certain terms present in the user interface input and / or based on other cues. Can be.

Each client computing device 1061 -N and automated assistant 120 facilitate communication via one or more memories for storage of data and software applications, one or more processors for accessing data and executing applications, and over a network. It may include other components. The operations performed by one or more client computing devices 1061 -N and / or automated assistant 120 may be distributed across multiple computer systems. Automated assistant 120 may be implemented as computer programs running on one or more computers, for example, in one or more locations coupled to each other via a network.

Automated assistant 120 may include term normalization engine 122, alternate input engine 124, scoring engine 126, one or more message generation engine 128, and chatspeak measurement engine 130. ) May be included. In some implementations, one or more of the engines 122, 124, 126, 128 and / or 130 can be omitted. In some implementations, all or aspects of one or more engines 122, 124, 126, 128, and / or 130 can be combined. In some implementations, one or more engines 122, 124, 126, 128, and / or 130 can be implemented as a component separate from automated assistant 120. For example, one or more engines 122, 124, 126, 128 and / or 130 or any operative portion thereof may be implemented in a component executed by one of the client computing devices 1061 -N.

In FIG. 1, the automated assistant 120 is shown in communication with the normalization database 152, the language model database 154, the response resource database 156, and the Internet language measurement database 158. Databases 152, 154, 156, and / or 158 are described in greater detail below, local to one of the client computing devices 106I, local to the automated assistant 120, and / or client computing devices 1061. -N) and / or on one or more non-transitory computer readable media, such as one or more media away from automated assistant 120. In many implementations, one or more databases 152, 154, 156 and / or 158 can be combined.

The term normalization engine 122 determines the normalization of various potential Internet term terms in the received natural language input. For example, the term normalization engine 122 may include normalization of "see" for term "c", normalization of "thanks" for term "thx", "review" and "rearview" for term "rvw", and the like. Normalizations can be determined. The term normalization engine 122 may use one or more mappings, rules, and / or other criteria of the normalization database 152 to generate normalizations. For example, the canonicalization database 152 may include mapping of specific Internet terminology terms to one or more corresponding normalizations. For example, the term "c" may map to a single normalization (normalization) of "see", and the term "rvw" may map to several normalizations such as "review", "rearview", and so forth. Also, for example, normalization database 152 may include multiple consecutive instances of the same character, and with terms that are not valid normalized terms, instances of that character may be removed until a valid normalized term is achieved. You can define a rule that says you can. For example, the rule normalizes "helloooooooo" to "hello", normalizes "yessss" to "yes", normalizes "yippeeeeeeeee" to "yippee", and "cooooolllN" to "cool" and "col "Can be used to normalize. In some implementations, the term normalization engine 122 may construct mappings and / or rules when determining normalization. For example, engine 122 may first dedupe "wazzaaaaa" to "wazza" and normalize "wazzaaaaa" to "what's up" by identifying the mapping between "wazza" and "what's up". . When de-duplicating "wazzaaaaa" to "what's up", engine 122 includes multiple consecutive instances of the same character, and is a term that is not a valid normalized term, such as a recognized term (e.g., Internet terminology terms). And terms recognized in the mapping between and normalized terms) can be relied upon to rule that instances of that character can be removed.

Substitution input engine 124 generates candidate substitution inputs for the received natural language input and generates each of the candidate substitution inputs by replacing one or more terms of the received natural language input with one of its normalizations, if any. For example, for a received natural language input of "tx 4 rvw", the term normalization engine 122 normalizes "tx" to "thanks", normalizes "4" to "for", and "rvw" to You can normalize with "review" and "rearview". Based on this normalization, the alternative input engine may include "thanks 4 the review", "" tx for the review "," thanks 4 the rearview "," thanks for the rearview "," thanks for the review "," "tx Candidate replacement inputs of "4 the review", "tx 4 the rearview".

The scoring engine 126 generates a score for each of the alternate inputs generated by the alternate input engine 124 and optionally generates a score for the natural language input for which the alternate inputs were generated. The scoring engine 126 uses the score to select at least one of them as the selected input to be used by the message generation engine 128 when generating a response to the received natural language input.

In various implementations, scoring engine 126 uses language model 154 to determine the score. In some of these implementations, the language model 154 is an n-gram trained on a corpus of other content and / or messages that include “normalized” (ie, non-internet) terms. gram) language model. For example, "see", "you" and "later" are more likely to occur than the terms "see,", "u" and "I8r" (used to train the language model 154). The language model 154 may be used to determine a score for “see you later” in the corpus). Compare this to the natural language input of "thx audio company" and the alternate input of "thanks audio company". In this situation, language model 154 may be used to determine that "thx", "company" and "audio" are more likely to occur more often than "thanks", "audio" and "company" (and thus better score for "thx audio company"). In some implementations, the scoring engine 126 presents the user with the term and / or the previous input that the user provided to the automated assistant 120 for a given input (received natural language input and / or alternate input). The terms of the input are used to determine a score for a given input based on the language model 154 without referring to the terms of the previous output provided by the automated assistant 120. In some other implementations, the scoring engine 126 may use one or more of the terms of the previous input (eg, immediately before) and / or the previous output (eg, immediately before).

As noted above, in some implementations, the language model 154 can be learned based on a corpus of other content and / or messages that include “normalized” terms. For example, the corpus of messages may include messages exchanged between two or more human users, and includes Internet language terms of messages converted to a normalized version by one or more human reviewers and / or automated systems. can do. For example, in a message containing "thanks 4 your help", the term "4" can be normalized to "for", and a message containing "for" instead of "4" may be used to train the language model 154. Are utilized. On the other hand, in a message containing "4 x 4 trucks", two occurrences of the term "4" are not replaced with "for", and a message containing "4 x 4 trucks" is used to train the language model 154. Can be used. In other words, in such implementations, the language model 154 can be learned based on a message having potential Internet language terms replaced with a normalized version when the terms are used in a fashioned manner. Potential Internet terminology terms are retained when used in an Internet language manner. Additions and / or alternatives to train the language model 154 in various implementations such as web pages, news articles, academic papers, patent pending publications, user internet comments provided for commenting on Internet content such as videos, and the like. This can be used.

In some implementations of determining a score, scoring engine 126 may include one or more response content scores from one or more message generation engines 128 and / or one or more rewrite scores from a rewrite engine (not shown). Additionally and / or alternatively use one or more additional scores such as In some of these implementations, scoring engine 126 may generate message generation engine (s) 128 (and / or other additional scores) when there is not enough deviation between two or more scores determined based on language model 154. Use the response content score (s) from the. In some other implementations, the scoring engine 126 always responds to the content content score (s) (and / or other additional scores) from the message generation engine (s) 128 as part of the score determination for one or more inputs being analyzed. May be used, and optionally using the response content score (s) in combination with the score (s) determined based on the language model 154.

In general, the response content score provided by one of the message generation engines 128 for an input (original natural language input or alternate input) is based on how good the response generated by that engine 128 is based on that input. (good) ". For example, one of the engines 128 may provide a “chat response” and may provide a response content score that indicates a good response if there is a scripted response that is strongly correlated to the input. For example, response resource database 156 may include one or more scripted chat responses, each mapped to one or more input parameters indicating when the scripted response is appropriate, with a response content score for a given input being predetermined. Can be determined based on a comparison of the input of the input and the input parameter (s). Also, for example, one of the engines 128 may provide content from “search results” such that at least one search result for a given input is a high quality search result and / or a particular “type” (eg In the case of "one box result", "knowledge card", and "curated response", a response content score may be provided for a predetermined input indicating a good response.

As an example, assume a received natural language input of "thx audio" and assume that alternate input engine 124 has generated a substitute input of "thanks audio". Also assume that scoring engine 126 generates scores for inputs based on language model 154, but these scores are so close to each other that one cannot guarantee to choose one (e.g., a better score). Does not meet the threshold for other scores). In this example, one of the message generation engines 128 can provide content from "search results" and for "thx audio" which represents much better response than the individual response content score for "thanks audio". Provide a response content score. The scoring engine 126 may select "thx audio" over "thanks audio" based at least in part on these response content scores.

As another example, assume a received natural language input of "thx 4 assistance" and assume that the replacement input engine 124 has generated a replacement input of "thanks for assistance". It is also assumed that scoring engine 126 generates scores for these inputs based on language model 154, but these scores are so close to each other that one cannot choose one over the other. In this example, one of the message generation engines 128 can provide content from the scripted response and have a response resource with input parameters of "thank for [help / assistance / input / info / information / etc.". May include a scripted response of "you're welcome" associated with database 156, where the parentheses "[]" and terms in parentheses represent a plurality of terms, one of which pairs with "thanks for the" Can be achieved. Based on the degree of strong conformity between the input parameters of the scripted responses "thanks for assistance" and "you're welcome", the message generation engine 128 scores the response content for "thanks for assistance". Can be provided as indicating a response that is much better than the individual response content score from its engine 128 for "thx 4 assistance." The scoring engine 126 may select "thanks for assistance" rather than "thx 4 assistance" based at least in part on these response content scores.

In general, a rewrite score provided by the rewriting engine for an input (original natural language input or alternate input) indicates how the rewriting of that input is "good", and the rewriting is based on the input and is optional Generated by the rewrite engine based on one or more previous inputs (or selected alternate inputs of previous inputs). How good a rewrite may be based on the nature of the rewrite (eg, syntactically correct) and / or the nature of the search result (s) responsive to the rewrite. For example, suppose a candidate alternative input including "what's the weather on Saturday", the current input of "what abt sun", and "what about Sunday" is the previous input. The rewrite of "what abt" based on the previous input may be "what's the weather on the Sun" and the rewrite of "what about Sunday" may be "what's the weather on Sunday". For example, based on one or more attributes of the search results responsive to the rewrite, the rewrite score for "what about Sunday" may be better than the rewrite score for "what about sun".

Regardless of the technique (s) utilized, scoring engine 126 may provide the selected input to message generation engine (s) 128, which responds to the selected input using response resource database 156. Response content may be generated. In some implementations, message generation engine (s) 128 can already generate the response content before receiving the selected input. For example, the response content may have been generated during generation of the response content score as described above. The generated response content is provided to the message exchange client 1071 -N that provided natural language input and provided for presentation to the user as a response to the natural language input. For example, the response content may be integrated by a message exchange client into a message exchange thread containing natural language input.

The Internet-facing measurement engine 130 determines the Internet-specific language specific value (s) for one or more users based on past usage of the Internet-facing words in the plurality of messages that contributed to the message exchange threads by one or more users. The Internet-facing measurement engine 130 stores the determined Internet-facing measurement (s) in the Internet-facing measurement (measurements) database 158 for use by one or more message generation engines 128. The message generation engine 128 includes in the response content it generates in response to the user's natural language input (whether or not the natural language input itself contains the Internet language terms), instead of their normalized terms, the Internet term term (s) When determining whether or not to include a user may use the Internet measurement measure (s) applicable to the user.

In some implementations, the internet pedometer measurement engine 130 determines the internet pedometer measurement based on the count of instances of the messages in which the internet is used. For example, the Internet-facing measurement engine 130 (a user or user as a function of the count of messages (of a user group's users) containing any Internet-speaker compared to the count of messages that do not contain any Internet-speaker). Determine "general" Internet language measures for the group. Also, for example, the Internet-facing measurement engine 130 may additionally and / or alternatively function as a function of the count of terms in a message (of a group of users) that are Internet-term terms compared to the total count of terms in a message. Determine "normal" Internet language measurements (for users or groups of users).

As one example, past instances of a pandemic translation for a user may be stored, and a pandemic may be determined based on the amount of these instances (e.g., based on low count, frequency, and / or other functions of the amount). Can be. For example, if a user consistently uses an Internet word in a message provided to the automated assistant, the Internet word measure for that user may indicate the user's desire to receive a response from the automated assistant that includes the word. In addition, the automated assistant may provide at least some responses that include the Internet Vocabulary term instead of the normalized version of the term. Internet term measurements may be "general" Internet term measurements or may be applied to one or more specific groups of Internet term terms. For example, a user's catchwords measure for "thx" / "thanks" might have the automated assistant replace "thanks" with "thx" in response to that user, while users for "gr8 / great" Internet measure of 's may not automatically replace "great" with "great" in response to the user.

In some implementations, the Internet language measurement may vary depending on the given user. For example, it may be determined based at least in part on instances of the Internet language in messages formulated by the user, such as instances recognized in accordance with the techniques described herein. In some of these implementations, the Internet language measurement may be specific to one or more devices of the user (eg, the user may be more likely to use the Internet language on a mobile computing device versus a “desktop” computing device). . In some implementations, the internet measure may be based on the internet instances of the plurality of users, and the internet measure may be based on a group of users, such as a user having one or more attributes in common (eg, geographic area, age range). And / or may be specific to the type of computing device used and / or the type of message exchange client used (e.g., message exchange client A may be more than client B among multiple users). May have a frequency of Internet traffic). In some implementations, a given user may cause the automated assistant 120 to include the internet language in the output provided to the user, such as the internet language that includes an internet language term associated with at least threshold internet language measurement for one or more groups of users. You may be offered the option to request it explicitly. For example, a user may choose to receive an Internet language term in the output, where the Internet term is popular among people in a geographic area (eg, critical internet measure and at least for critical group of users within a geographic area). Associated Internet terminology).

In some implementations, the plurality of Internet language measurements may be used by one of the message generation engines 128 when determining whether to use one or more Internet terminology terms instead of the normalized version of these terms. For example, the message generation engine 128 outputs to be presented to the user based on both the user's general internet measurement and the Internet measurement (specific to the user or group of users) specified in "c / see". You can decide whether to use "c" instead of "see" in. For example, assume that a user's general internet vocab measurement represents the "medium" degree of user's use of the internet. If the internet measure measured in "c / see" indicates a high level of internet usage in association with that term, "c" may be used in place of see "in the generated output, while" c / see " A "see" may be used in the generated output if a specified internet measure indicates a low level of usage of the word in association with the term.

In FIG. 2, additional descriptions of the various components of the automated assistant 120 are provided. As shown in FIG. 2, a user provides natural language input 201 to message exchange client 107i using one or more user interface input devices of computing device 106i, which message exchange client 201 ) Is sent to the term normalization engine 122. Natural language input 201 may be free input, for example, a typed input provided by a user through a physical or virtual keyboard of a computing device or a voice input provided by a user to a microphone of a computing device. In implementations where natural language input 201 has been entered, this may optionally be indicated by text input by the computing device and / or by a component of automated assistant 120.

In an embodiment, the natural language input 201 may be “c u I8r”, which is also shown in FIG. 2 in parentheses to facilitate understanding of the embodiment.

The term normalization engine 122 generates one or more normalizations for one or more of the terms of the natural language input 201, optionally based on one or more mappings, rules, and / or other criteria of the normalization database 152. The term normalization engine 122 provides the normalized term 202 to the alternate input engine 124.

Continuing the embodiment, the term normalization engine 122 generates a normalized term 202, where "c" is normalized to "see", "u" is normalized to "you", and "I8r" is Normalized to "later". The normalizations are shown in parentheses in FIG.

Alternate input engine 124 generates alternate inputs 203 of natural language input 201 that each include at least one of normalized terms 202 that replaces one of the terms of natural language input 201.

Continuing with an embodiment, alternative inputs 203 generated based on natural language input 201 of “cu I8r” and normalized terms of “see”, “you” and “later” are “see u I8r,” c you I8r "," cu later "," see you I8r "," see u later "," c you later ", and" see you later ", which are also shown in parentheses in FIG.

The scoring engine 126 generates a score for each of the alternate inputs and optionally for the natural language input 201 and based on the score, select one or more (eg only one) of the selected inputs 204. Select as. In various implementations, scoring engine 126 uses language model 154 to determine scores. In some of these implementations, language model 154 learns about a corpus of other content and / or messages that includes, is restricted, or is governed by "normalized" (ie, non-Internet language terms) terms. N-gram language model.

Continuing with the embodiment, the language model 154 is more likely to occur at the same time (see "" you ") at the same time (in the corpus used to train the language model 154) than the terms of the other alternative inputs 203. Can be used to determine a score for "see you later" representing "and" later ".

In some implementations of determining scores, scoring engine 126 additionally and / or alternatively uses one or more response content scores from one or more message generation engines 128. In some of these implementations, the scoring engine 126 may not be able to obtain a message from the message generation engine (s) 128 when there is not enough variation between two or more of the scores determined based on the language model 154. Use the response content score. In general, the response content score provided by one of the message generation engines 128 for an input (original natural language input or alternate input) is based on how "good" the response generated by that engine 128 is based on that input. Hanji. For example, one of the engines 126 may provide a “chat response” and may provide a response content score that indicates a good response if there is a scripted response that is strongly correlated to the input. Also, for example, one of the engines 128 may provide content from “search results” such that at least one search result for the input is a high quality search result and / or a particular “type” (eg “ one box result "," knowledge card ", and" curated response "), a response content score indicating a good response may be provided.

Regardless of the technique (s) used, the scoring engine 126 selects the selected message in the message generation engine (s) 128 using the response resource database 156 to generate response content 205 responsive to the selected input. Provide input 204. The response content 205 is provided to the message exchange client 107i for an audible and / or visual presentation to the user as a response to the natural language input 201. For example, the response content 205 may be integrated into the message exchange thread by the message exchange client 1071 to display simultaneously with the natural language input 201 of the message exchange thread.

Also shown in FIG. 2 is the Internet Vocabulary Measurement Engine 130 and the Internet Vocabulary Measurement (Measures) Database 156. In some implementations, in response to determining the selected input 204 comprising "translation" of one or more Internet term terms into a normalized term, the scoring engine 126 sends the translation engine to the Internet measurement engine 130. It may provide an indication of occurrence and / or provide an indication of translated Internet language terms and / or normalized terms. As described herein, in some implementations, the indication and / or additional indication may be a “Internet Vocabulary measure” for one or more users (and optionally one or more specific Internet Vocabulary terms and / or normalized terms). ) Can be used to determine As described herein, the Internet Language Measurements and / or additional Internet Language Measurements may or may not include the Internet Terms term (s) in the response content 205 and / or the Internet Terms in the additional response content instead of the normalized terms. It may be used by the message generation engine (s) 128 to determine whether to include (s).

Continuing with an embodiment, the scoring engine 126 may provide the generalized measurement engine 130 with a general indication that a "translation" has occurred in which a natural language input including the trendy language has been converted into a substitute input. . The generic indication may be used by the Internet Measurement Engine 130 to determine general Internet Measurements for a user, for a group of users, and / or for other parameters. The scoring engine 126 additionally and / or alternatively displays the representation of the Internet Vocabulary and Internet Vocabulary Phrases and / or selected normalized versions of the Internet Vocabulary Terms in the Internet Vocabulary Terms and / or Natural Language Input. 130 may be provided. For example, if the input 201 is "cu I8r" and the selected input 204 is "see you later", the scoring engine 126 will normalize the Internet of "c" to "see" and " indication (s) that the u-word of "u" has been normalized to "you", the u-word of "I8r" has been normalized to "later", and / or the word "cu I8r" has been normalized to "see you later". May be provided to the Internet trend measurement engine 130. These indications may be used by the Internet Measurement Engine 130 to generate one or more Internet Measurements specific to these terms and / or phrases, and via the Message Exchange Client 107i, through other Message Exchange Clients. May be further based on other translation instances of these terms and / or phrases by the user through, and / or other applications.

The message generation engine (s) 128 may use the monetary measurement (s) of the database 156 in determining whether to include the monetary term (s) in the response content 205 and / or additional response content. . Continuing the embodiment, assuming that the selected input 204 is "see you later", one of the message generation engines 128 is from the response resource database 156 to "see you then" for the selected input 204. Determine the scripted response of the. Based on one or more Internet language measurement (s) in database 156 applicable to the user of message exchange client 107i, message generation engine 128 replaces "see" with "c" and replaces "you". We can substitute "u" and provide the output of "cu then".

With reference to FIGS. 3 and 4, additional descriptions are provided for the various components and techniques described herein. 3 and 4 also show display screen 140 and FIG. 1 displaying an example of a conversation that may occur between a user of client computing device 106i and automated assistant 120 in accordance with an implementation described herein. Client computing device 106i is shown. Display screen 140 also includes system interface elements 381, 382, 383 that can interact with a user to cause computing device 106i to perform one or more operations. The display screen 140 further includes a text response interface element 384 in which the user can select to generate natural language input through a virtual keyboard and a voice response interface element 385 in which the user can select to generate natural language input through a microphone. Include.

3 includes natural language input 380A-D submitted by a user through one or more user interface input devices of the client computing device 106i. 3 further includes responses 390A-390D generated by automated assistant 120.

Input 380A of "Helloooooooooooooo" may be normalized to "hello" by the term normalization engine 122. Alternate input engine 124 may use "hello" as an alternate input. The scoring engine 126 generates a first score for "Helloooooooooooooo", a second score for "hello", selects "hello" as the selected input based on the scores, and generates a "hello" message. May be provided to the engines 128. One of the message generation engines 128 may generate a response 390A based on a chat response that is mapped or otherwise associated with "hello" in the response resource database 156.

“Gr8” of input 380B may be normalized to “great” of input 380B, and “thx” of input 380B may be normalized to “thanks” by the term normalization engine 122. The alternate input engine 124 may generate alternate inputs of "gr8 thanks", "great thx", and "great thanks". Scoring engine 126 may generate a score for each of input 380B and alternate inputs. In some implementations, a score for “great thanks” can be generated based at least in part on the response content score as described herein, wherein the response content score can optionally be sent to the previous response 390A and / or previous. It may be generated by one of the message generation engines 128 in consideration of the input 380A. The scoring engine 126 selects "great thanks" as the selected input to provide to the message generation engine 128. One of the message generation engines 128 may generate a response 390B based on a chat response that is mapped or otherwise associated with “Good to hear” in the response resource database 156.

Input 380C is "c e o of numbers company?" The term normalization engine 122 may normalize "c" to "see", "e" to "enemy", and "o" to "oh" and "over". The alternate input engine 124 is "see e o of numbers company?", "See enemy over numbers company?" Alternative inputs can be generated. Scoring engine 126 may generate a score for each of input 380C and alternate inputs. The scoring engine 126 may select, based on the scores, input 380C of " c o of numbers company? &Quot; and provide it to the message generation engine (s) 128. One of the message generation engines 128 may generate a response 390C based on the search result returned for input 380C. The response 390C includes the text "Here's what I found:" indicating that the content responds to input 380C, and is a "card" focused on the fictional CEO of the fictional "Numbers Company". ) ". The card includes a selectable option to view the CEO image, CEO name, and additional information that, when selected (in the message exchange client 107i or a separate application), can provide the user with more information related to the CEO. . The input 380D of “cu I8r” may be translated by the automated assistant 120 to “see you later” and may be used to provide a scripted response 390D of “Later”.

4 includes natural language input 480A-D submitted by a user through one or more user interface input devices of the client computing device 106i. 4 further includes responses 490A-490D generated by automated assistant 120. The example of FIG. 4 shows, among other things, the Internet of Things measurement (s) associated with a user of the client computing device 106i for selectively incorporating the Internet of Things into the generated responses 490A-490D. Show how you can use).

The automated assistant 120 generates a response 490A of "Gr8 thx-and you?" In response to the input 480A of "how r u today?". The response 490A includes the Internet language terms "Gr8" and "thx" and may include the Internet language term based on the Internet language measurement as described herein. In some implementations, one of the message generation engines responds based on being represented as a scripted Internet Language Response to input 480A (or a translation of input 480A determined as described herein). 490A can be determined. In such implementations, the message generation engine 128 selects the response 490A based on the Internet pedometer measurement of the user who submitted the input 480A that satisfies the threshold and based on what is represented as a “Internet pod” response. Can be. In some implementations, one of the message generation engines 128 first determines “Great thanks-and you?” In response to the input 480A (or translates the determined input 480A as described herein). ) May then determine the response 490A based on changing "Great" to "Gr8" and "thanks" to "thx" but not "you" to "u". For example, a user's internet measure may indicate that the user typically uses "Gr8" and "thx" as an Internet fashion, but may indicate that "u" is rarely (or not) used as an internet fashion. have.

Input 480B of " gr8 as well " is responded with a scripted response 490B of " Good to hear " that does not include an Internet Term. For example, one of the message generation engines 128 may return a response 490B that is represented as a scripted Internet Explorer response to an input 480B (or a translation of the input 480B determined as described herein). You can decide.

Input 480C, "Who's the dog in Acme Movie", is answered in response 490C. One of the message generation engines 128 may generate a response 490C based on the search result returned for input 490C. Response 490C indicates "Here's" to indicate that the content is in response to message 480C and includes a "card" that focuses on a fictional dog acting on the fictional movie "Acme Movie." Include the text what I found: ". The card contains the dog's image, the dog's "official" name (WALTER P.), and additional information about the dog. The additional information includes the name "Great" of the dog playing the role in the virtual Acme movie. It is also noted that the term "Great" was not translated in this additional information despite the response of 490A. This applies to rules that do not use Internet terminology instead of the term of a particular response, such as, for example, the term of a response generated based on a search result (eg, compared to a response generated based on a scripted response). Can be based.

An input 480D of "c u I8r" is responded with a response 490D of "I8r" that includes an Internet idiom and may be generated according to the techniques described herein.

FIG. 5 illustrates the use of the normalized form to receive a natural language input including an Internet language, convert the Internet language into a normalized form, and generate response content responsive to the natural language input, and to respond to the natural language input. A flow diagram illustrating an example method 500 of providing response content to present is presented. For convenience, the operation of the flowchart is described with reference to the system performing the operation. The system may include various components of various computer systems, such as automated assistant 120. In addition, although the operation of method 500 is shown in a particular order, it is not intended to be limiting. One or more actions may be rearranged, omitted, or added.

In block 550, the system is formulated by the user through one or more user interface input devices, and includes a message exchange thread (e.g., a thread between the user and the automated assistant and / or a user, automatic assistant and one or more additions). A natural language input provided by the user in a thread between users).

At block 552, the system identifies one or more normalizations for each of the one or more terms of the natural language input of block 550.

At block 554, the system generates candidate replacement inputs for the natural language input of block 550 based on the normalization of block 554.

At block 556, the system determines a score for the natural language input of block 554, and determines a score for each of the candidate replacement inputs determined at block 554. In some implementations, in determining the scores, the system determines one or more scores based on the language model at block 556A and / or one of the scores based on the response content score at block 556B. Determine the above.

At block 558, the system selects one or more (eg, only one) input based on the score. In particular, the system selects one or more of the candidate replacement inputs of block 554 and / or the natural language input of block 550 based on the scores for these inputs.

At block 560, the system determines response content based on the input (s) selected at block 558. For example, the system can determine response content that is a chat response or response content that is a search result response.

At block 562, the system integrates the response content of block 560 into the message exchange thread as a response to the natural language input.

FIG. 6 is a natural language output for determining an Internet spoken word measurement for a user, receiving a natural language input from the user, determining a natural language output based on the internet translated word measurement and the natural language input, and displaying it to the user in response to the natural language input. Provides a flow diagram illustrating a method 600 according to an embodiment of the present invention to provide a. For convenience, the operation of the flowchart is described with reference to a system that performs the operation. The system may include various components of various computer systems, such as automated assistant 120. In addition, although the operation of method 600 is shown in a particular order, it is not limited thereto. One or more actions may be rearranged, omitted, or added.

In block 650, the system determines a monetary measure for one or more users based at least in part on the count of instances of the monetary words of messages that contribute to the message exchange thread by the one or more users.

In block 652, the system is formulated (presented) by the user via one or more user interface input devices, and an ongoing message exchange thread (e.g., a thread between the user and the automated assistant and / or the user, automated). A thread between an assistant and one or more additional users).

At block 654, the system determines a natural language incorporating one or more Internet term terms based on the internet language measurement determined at block 650 and based on the natural language input at block 652. In some implementations, the system determines a natural language output that incorporates one or more Internet Vocabulary terms based on the Internet Vocabulary measurements determined at block 650 and based on one or more additional Internet Vocabulary measurements.

At block 656, the system integrates the response, including the natural language output of block 654, into an ongoing message exchange thread.

7 is a block diagram of an example computing device 710 that can optionally be used to perform one or more aspects of the techniques described herein. In some implementations, one or more client computing devices 1061 -N, automated assistant 120, and / or other component (s) may include one or more components of example computing device 710.

Computing device 710 typically includes at least one processor 714 that communicates with a number of peripheral devices through bus subsystem 712. Such peripheral devices may include, for example, storage subsystems including memory subsystem 725 and file storage subsystem 726, user interface output device 720, user interface input device 722, and network interface subsystem 716. System 724 may be included. The input and output devices allow user interaction with the computing device 710. Network interface subsystem 716 provides an interface to an external network and is coupled to a corresponding interface device of another computing device.

User interface input devices 722 may be pointing devices such as a keyboard, mouse, trackball, touch pad or graphics tablet, scanner, touch screen integrated into the display, voice recognition system, microphone and / or other types of input devices. It may include an audio input device. In general, the use of the term “input device” is intended to include all possible types of devices and methods for entering information on computing device 710 or a communication network.

User interface output devices 720 may include non-visual displays, such as a display subsystem, printer, fax machine, or audio output devices. The display subsystem may include a cathode ray tube (CRT), a flat panel device such as a liquid crystal display (LCD), a projection device, or other mechanism for generating a visible image. The display subsystem may also provide a non-visual display through an audio output device. In general, the use of the term “output device” is intended to include all possible types of devices and methods for outputting information from computing device 710 to a user or other machine or computing device.

Storage subsystem 724 stores programming and data structures that provide the functionality of some or all of the modules described herein. For example, storage subsystem 724 can include logic to perform selected aspects of the methods of FIGS. 5 and 6.

These software modules generally execute on the processor 714 alone or in combination with other processors. Memory 725 used in storage subsystem 724 includes main random access memory (RAM) 730 for storage of instructions and data during program execution and read only memory (ROM) 732 in which fixed instructions are stored. It may include a plurality of memories including a. File storage subsystem 726 may provide persistent storage for program and data files and may include a hard disk drive, a floppy disk drive with associated removable media, a CD-ROM drive, an optical drive, or a removable media cartridge. Can be. Modules that implement the functionality of certain implementations may be stored on the file storage subsystem 726 of the storage subsystem 724 or on another machine accessible by the processor (s) 714.

Bus subsystem 712 provides a mechanism that allows various components and subsystems of computing device 710 to communicate with each other as intended. Although bus subsystem 712 is schematically depicted as a single bus, other implementations of the bus subsystem may use multiple buses.

Computing device 710 may be of various types, including a workstation, server, computing cluster, blade server, server farm, or any other data processing system or computing device. Due to the ever-changing nature of computers and networks, the description of computing device 710 shown in FIG. 7 is intended only as a specific example for describing some implementations. Many other configurations of the computing device 710 are capable of having more or fewer components than the computing device shown in FIG. 7.

In situations where the systems described herein may collect or use personal information about users, the user may be interested in the systems, programs or functions described herein for user information (e.g., a user's social network, Information about social actions or activities, occupations, user preferences, or the user's current location) and when the user sends content or communications from the server. Control may be provided. In addition, certain data may be processed in one or more ways prior to being stored or used to remove personally identifiable information. For example, the user's identity may be determined so that personally identifiable information about the user cannot be determined, or by generalizing the user's geographic location from which the location information was obtained (e.g., city, postal code, or state level). It can be processed so that a specific location of cannot be determined. Thus, a user can control what information is collected about the user, how that information is used, and what information is provided to the user.

While some embodiments have been described and illustrated herein, various other means and / or structures can be used to perform the functions and / or obtain the results and / or one or more advantages described herein, such variations and Modifications are considered to be within the scope of the embodiments described herein. More generally, all of the parameters, dimensions, materials, and configurations described herein are exemplary, and actual parameters, dimensions, materials, and / or configurations are dependent upon the particular application or use for which the teaching is used. something to do. Those skilled in the art will recognize, or be able to ascertain many equivalents to the specific embodiments described herein using routine experiments only. Accordingly, it is to be understood that the foregoing embodiments are merely examples, and that within the scope of the appended claims and their equivalents, the embodiments may be practiced otherwise than as specifically described and claimed. Embodiments of the present disclosure relate to each individual feature, system, article, material, kit, and / or method described herein. In addition, all combinations of two or more such features, systems, articles, materials, kits, and / or methods are included within the scope of the present disclosure unless such features, systems, articles, materials, kits, and / or methods are inconsistent with each other. .

Claims (20)

A method performed by one or more processors,
Determining chatspeak measurements for one or more users based on the plurality of messages contributing to message exchange threads by the one or more users;
Receiving natural language input submitted by at least one of the users as part of an ongoing message exchange thread between the user and the automated assistant, wherein the natural language input is via the user interface input device of the user's computing device. Represented by;
Generating a natural language output for inclusion in a response of the automated assistant to the natural language input, wherein generating the natural language output includes incorporating one or more Internet terminology terms into the natural language output based on the internet language measurement. Comprising; And
Including the response of the automated assistant in the message exchange thread, wherein the response includes the natural language output with the one or more Internet terminology terms. . The method of claim 1,
Incorporating the response into the message exchange thread includes inserting the response into a transcript of the message exchange thread displayed in a graphical user interface of a message exchange client operating on the computing device of the user. Characterized by a method performed by one or more processors. The method of claim 1,
The Internet term measurement is specific to a group of one or more Internet term terms that include at least the Internet term terms included in the natural language output, and incorporating the one or more Internet term terms into the natural language output includes: Performing the one or more processors based on the Internet language measurement specific to the group. The method of claim 1,
Incorporating the Internet terminology into the natural language output,
Identifying an initial natural language output responsive to the natural language input;
Determining whether one or more predetermined terms of the initial natural language output map to the Internet terminology terms; And
Replacing the predetermined terms of the initial natural language output with the Internet terminology terms. The method of claim 1,
Determining additional internet language measurement for the one or more users, wherein generating the natural language output is based on the internet language measurement and based on the additional internet language measurement; Incorporating an ideal into the natural language output. The method of claim 5,
The internet term measurement is specific to the user and is applicable to a plurality of internet term terms including one or more additional internet term terms that are not integrated into the natural language output,
And wherein said additional internet measure is not specific to said user and is specific to at least one of said internet terminology incorporated into said natural language output. 10. The method of claim 1, wherein the Internet spoken measurement is specific to the user. The method of claim 7, wherein
And wherein said message exchange threads comprise at least automated assistant message exchange threads of the past between said user and said automated assistant. The method of claim 8,
Determining the internet spoken measurement is based on a count of the instances of the internet spoken used in the messages, and the method further comprises determining one of the instances on which the count is based;
Determining one of the instances,
Receiving a predetermined one of the messages in one of the past automated assistant message exchange threads, wherein the predetermined message is represented by a user and includes at least a first term and a second term;
For at least the first term of the natural language input, identifying a first normalization of the first term;
Generating a candidate substitute input for the natural language input, the generating step replacing the first term with the first normalization and including the second normalization or second term of the second term in the candidate substitute input. Including making;
Determining a first score for the natural language input;
Determining a second score for the candidate substitute input;
Selecting the candidate substitute input based on the first score and the second score;
Determining response content based on the candidate substitute input;
Incorporating the response content into one of the past automated assistant message exchange threads in response to the natural language input by the automated assistant, wherein incorporating the response content comprises selecting the candidate alternate input. Based; And
And storing an indication of one of the instances based on selecting the candidate replacement input. The method of claim 9,
Determining the first score for the natural language input is based on applying at least the first term and the second term to a language model,
Determining the second score for the candidate substitute input is based on applying at least one of a second term or a second normalization and the first normalization to the language model. Method performed by. The method of claim 9,
The internet term measurement is specific to the user, the internet term terms,
Storing an indication of one of the instances based on selecting the candidate substitute input further comprises storing an association of at least the first term or an indication for the first normalization. Method performed by the above processor. The method of claim 1,
Determining the Internet fashionable measurement,
The Internet word is selected from the group consisting of a count of instances used in the messages, a quality of one or more Internet term terms used in the messages, a commonality of the Internet word terms, and a consistency of the use of the Internet word terms. Method performed by one or more processors, characterized in that it is based on at least one. As a system,
Memory for storing instructions;
One or more processors executing the instructions stored in the memory, the instructions causing the one or more processors to perform the operations,
Determining a chatspeak measurement for one or more users based on the plurality of messages contributing to message exchange threads by the one or more users;
Receiving natural language input submitted by at least one of the users as part of an ongoing message exchange thread between the user and the automated assistant, the natural language input being via the user interface input device of the user's computing device. Represented by;
Generating a natural language output for inclusion in the response of the automated assistant to the natural language input, wherein generating the natural language output includes incorporating one or more Internet terminology into the natural language output based on the internet language measurement; Comprising; And
Including the response of the automated assistant in the message exchange thread, the response comprising the natural language output having the one or more Internet terminology terms. The method of claim 13,
In incorporating the response into the message exchange thread, the one or more processors insert the response into a transcript of the message exchange thread displayed in a graphical user interface of a message exchange client operating on the user's computing device. System characterized in that. The method of claim 13,
The Internet term measurement is specific to a group of one or more Internet term terms that include at least the Internet term terms included in the natural language output, and incorporating the one or more Internet term terms into the natural language output includes: A system based on the internet language measurement specified in the group that includes. The method of claim 13,
In incorporating the Internet terminology into the natural language output, the one or more processors include:
Identify an initial natural language output responsive to the natural language input;
Determine whether one or more predetermined terms of the initial natural language output map to the Internet terminology terms; And
And replace the predetermined terms of the initial natural language output with the Internet terminology. The method of claim 13,
In executing the instructions, the one or more processors further determine additional Internet language measurements for the one or more users,
In generating the natural language output, the one or more processors incorporate one or more of the Internet language terms into the natural language output based on the internet language measurement and based on the additional internet language measurement. The method of claim 13,
The internet spoken measurement is specific to the user,
And the message exchange threads comprise at least automated assistant message exchange threads of the past between the user and the automated assistant. The method of claim 18,
In determining the internet pedometer measurement, the one or more processors determine the internet pedometer measurement based on a count of instances in which the internet spoken word was used in the messages;
In executing the instructions, the one or more processors further determine one of the instances on which the count is based,
In determining one of the instances, the one or more processors,
Receive a predetermined one of the messages in one of the past automated assistant message exchange threads, wherein the predetermined message is represented by a user and includes at least a first term and a second term;
For at least the first term of the natural language input, identify a first normalization of the first term;
Generate a candidate substitute input for the natural language input;
Determine a first score for the natural language input;
Determine a second score for the candidate substitute input;
Select the candidate substitute input based on the first score and the second score;
Determine response content based on the candidate substitute input;
Based on selecting the content, incorporating the response content into one of the past automated assistant message exchange threads in response to the natural language input by the automated assistant;
Store an indication of one of the instances based on selecting the candidate substitute input. The method of claim 13,
In determining the Internet of Things measurement, the one or more processors are further configured to determine a count of instances in which the Internet spoken word was used in the messages, the quality of the one or more Internet Vocabulary terms used in the messages, And determine the internet measure based on at least one selected from the group consisting of commonality and consistency of use of the internet terminology.

Images